저자: 젠슨황 · 외국인
요약 보기
엔비디아 CEO 젠슨 황은 AI 인프라 투자가 2000년 닷컴 버블과 다르다고 주장하며, 하이퍼스케일러 사업 규모(약 2.5조 달러)와 연간 자본지출(약 5000억 달러)을 근거로 제시한다. 이 전환은 CPU에서 GPU 기반 생성형 AI로 이행하는 초기 단계이며 엔비디아 AI 인프라 사업은 수천억 달러 규모로 여전히 초기라고 말한다. 토큰 생성이 이제 수익성을 갖췄다고 강조하고, 기업용 AI(Cursor 등)의 확산을 근거로 든다. 엔비디아가 xAI 등 AI 기업에 투자하는 것에 대해 후회 없이 더 투자하지 못한 것이 아쉽다고 말한다.
다듬은 글
Let's talk about vendor financing because that has been something that has raised a lot of questions on Wall Street since you cut this deal with OpenAI. Uh yesterday it Bloomberg is reporting that you have a $2 billion in financing that you're going to be involved with with XAI to help them with these same stories.
But this idea of circular financing, your your customers can't afford to buy these chips yet, so you're going to help them out with money along the way. that leads some people to think back to what happened u with a with companies during uh the big buildup like a Lucent or Nortell in the early days.
Is this different? And if so, how? Well, first of all, XAI uh I'm super excited about the financing opportunity that they're doing. You know, if the only the only regret I have about XAI, we're an investor already. The the only regret I have is I didn't give him more money. You know, almost everything that Elon's part of, you really want to be part of as well.
and and uh he gave us the opportunity to invest in XAI. I'm just delighted by that. And so that's not that's an investment into a really great future company and I'm really excited about that. That's not venture venture uh that's not uh vendor financing per se, but what's going on what's going on in the world versus what happened in 2000 is just dramatically different.
You know back then as you recall there were pets.com hospitals.com and the the all of the internet companies combined was what 3040 billion dollars in size. If you look at the hyperscalers now that's where the first trunch of AI infrastructure is building. If you look at the AI the the hyperscalers have about $2.5 trillion of business that's already operating today.
that business, that $2 and half trillion dollar business and the capex that goes underneath that is about call it 500 billion dollars. That transition from a classical CPUbased computing to now generative AI computing powered by GPUs, that transition is just starting. So we're we've got to build into half a trillion dollars worth of capacity infrastructure that's already naturally growing by itself.
And we're in the beginning phases of that. We're, if you just look at Nvidia's AI infrastructure business, you know, call it a couple of hundred billion dollars so far. You we're a couple hundred billion dollars into a multi-t trillion dollar buildout. So that's number one. The second part of it that's really unique is that we have a new generation of AI companies.
the new AI companies like OpenAI and Anthropic and XAI and and companies that are that are well you know thinking machine labs from Mera and Ilas Susver SSI and Misha's company reflection and I mean there's a whole bunch of amazing AI AI model builders now this generation of AI model builders what's happened in the last several months a transition happened that is really really important for the last several years they've been generating tokens tokens, you know, these AI tokens basically at a loss.
And the reason for that is because the early AI models weren't they were super interesting uh uh really captivated a lot of attention, but they weren't useful enough to pay for. The last several months, it's been very clear that the new technology is now reasoning. It's doing research. Before it answers a question, it goes on the web and studies other PDFs and websites.
It can now use tools, generate information for you, and it creates creates responses that are really useful. I use it every day to the point where now the tokens are profitable. The the question though is who is going to continue to pay for that buildout? Is it the big companies like uh Proctor and Gamble?
Is it a big is it consumers who are going to do this? I mean, my doctor showed me the AI that he's using and it's incredibly helpful, but there's still a question about whether he's going to pay for it or the company he works for is going to pay for it. Well, hopefully both. I think the uh there's the consumer part of it.
You know, a lot of OpenAI customers are consumers and they're paying for it. Um, but the thing that's really cool is that the enterprise AI buildout that's happening now, my my favorite enterprise AI c uh uh service is cursor. Cursor is an AI coder. And every one of our engineers 100% is now assisted by AI coders.
And our productivity has gone up incredibly. And so you're now seeing enterprise AI companies like Curser, Open Evidence, I love Lovable, um, all of these companies are some of the fastest growing companies in the world and they address enterprise and so enterprise AI is here. I wonder how you think about Jensen the ultimate destination uh, where all this is actually building toward.
Um I don't know there was a a post from David Conniey's a Sequoia partner this week saying look AGI is the only thing that can justify the volumes of capital spending right now artificial general intelligence at the same time many of the model builders and experts are pushing out the date where that maybe is going to be achievable.
Uh meantime you're talking about annual generations of chips. How fast did they depreciate? we're putting all this capital in these data centers that it's just not clear like when we get to the end or is it just a treadmill? We are going to have incredibly profitable and incredibly useful AIs long before AGI and for example right now cursor AI you know cursor cursor is a AI software coder is incredibly useful all of our engineers use it we have some 40,000 engineers uh almost every one of them are going to use it and they're they're loving it and so we're
they're using it instead of something else presumably right so these are replacing things yeah um they're not they're this is a brand new Yeah, remember AI unlike unlike previous technologies, previous technologies are tools that humans use. Excel is a tool that humans use. Um a web browser, it's a tool that humans use.
For the very first time, we have technology that can actually use tools by itself. And so cursor uses, you know, visual C++ and it it uh uh and now we have uh uh Gemini agents uh that are able to use the browser and browse browse for for uh uh groceries or you know destinations or book travel for you and and it can use it can use tools by itself.
So so this is really quite an extraordinary thing. This tool users the tool industry is a few trillion dollars. Yeah. Tool user industry is a hundred trillion dollars. which is the reason why everybody's so excited about the future of technology because it could augment labor. It can increase the productivity of labor and here at NVIDIA, you know, it's increased our product productivity tremendously.
You've already said you wish you could have bought more open AI and maybe invested more in XAI at this point. that implies that you don't think that this is all redundant because a lot of the players, I mean, if I read what Sam Alman says and the message he tries to convey, it's as if they're sprinting to try to stay ahead of everybody else because they think it's not going to be room for everybody necessarily who's trying to do something similar.
I think there's general intelligence and I think there's specialized intelligence. We love general when I hire engineers, I like them to be generally intelligent and that's that's a great thing. But once they come to Nvidia, we make them highly specialized intelligent so that they could build things that you know Nvidia needs.
And so I I think the idea of specialized intelligence versus generalized intelligence will continue to happen. And where the real value for enterprises and companies are is specialized intelligence and where the value for consumers general intelligence. Hey Jensen, let's talk a little bit about capital allocation and how you see that.
Just back to those same points that you wish you had more invested in open AI. you wish you had more invested um in in in these other companies that you're kind of taking stakes in like an XAI. You can do things with your money. You can choose to put it back into R&D. You probably don't need a ton with that right now.
You can choose to build the company. You can choose choose to make acquisitions. You can choose to give shareholders back big dividends. Or you can invest it yourself in other companies. And lately, what we've seen is a lot of investing in other companies. Do you have other deals like that planned in the works?
We're always looking for great startups to invest in. Uh, one of my favorite ones was Coreweave, right? My only regret is I didn't invest add that to the list enough. Yeah. And so in all of these all of these investments that we've made recently, um, we've made some really terrific investments and and largely my only regret is that we didn't invest more because these are really special companies and they're building they're part of our ecosystem building out the AI um, infrastructure for the world.
And AI is several things. AI is energy. AI's chips, the models, and the applications. And so you could see Nvidia and you could look at me working across that entire stack of ecosystems around the world. And um I we need more energy, we need more chips, we need better models and and more models. And we need a lot more applications.
Comments 0